ABCG5 and ABCG8 Are Obligate Heterodimers for Protein Trafficking and Biliary Cholesterol Excretion*

ABCG5 (G5) and ABCG8 (G8) are ATP-binding cassette (ABC) transporters that limit intestinal absorption and promote biliary excretion of neutral sterols. Mutations in either ABCG5 or ABCG8 result in an identical clinical phenotype, suggesting that these two half-transporters function as heterodimers. Expression of both G5 and G8 is required for either protein to be transported to the plasma membrane of cultured cells. In this paper we used immunofluorescence microscopy to confirm, in vivo, that G5 is localized to the apical membranes of mouse enterocytes and hepatocytes. Other ABC half-transporters function as homodimers or as heterodimers with other subfamily members. To determine whether G5 or G8 complex with other ABCG half-transporters, we co-expressed G1, G2, and G4 with either G5 or G8 in cultured cells. G1, G2, and G4 co-immunoprecipitated with G5, and G4 co-immunoprecipitated with G8, but the putative dimers were retained in the endoplasmic reticulum (ER). Adenovirus-mediated expression of either G5 or G8 in the liver of G5G8 null mice resulted in ER retention of the expressed proteins and no increase in biliary cholesterol. In contrast, co-expression of G5 and G8 resulted in transit of the proteins out of the ER and a 10-fold increase in biliary cholesterol concentration. Finally, adenoviral expression of G2 in the presence or absence of G5 or G8 failed to promote sterol excretion into bile. These experiments indicate that G5 and G8 function as obligate heterodimers to promote sterol excretion into bile.

ABCG5 (G5) 1 and ABCG8 (G8) are members of the large family of ATP-binding cassette (ABC) transporters that facilitate translocation of a wide variety of substrates across cellular membranes (1). Eukaryotic ABC transporters consist of two modules, a magnesium-dependent ATPase catalytic domain and a transmembrane domain containing 6 -12 membranespanning segments (2). ABC transporters are divided into halftransporters, which contain one ATPase domain and one membrane-spanning domain, and full-transporters, which contain paired modules in tandem (2). G5 and G8 are both half-transporters belonging to the G subfamily of ABC proteins. Like the other three members of the human G subfamily of ABC transporters (ABCG1, ABCG2, and ABCG4), the ATPase catalytic domains of G5 and G8 are located N-terminal to the transmembrane domain.
Mutations in either ABCG5 or ABCG8 cause sitosterolemia, an autosomal recessive disorder characterized by the accumulation of both plant-derived (primarily sitosterol) and animalderived (cholesterol) sterols in plasma and tissues (3)(4)(5). In mice, G5 and G8 limit the absorption of dietary sterols and promote the excretion of cholesterol into bile (6,7). In sitosterolemia there is a generalized increase in the absorption of dietary neutral sterols and a defect in the efflux of these sterols into bile (8 -10). These changes in sterol trafficking result in deposition of neutral sterols in skin as xanthomas and in coronary arteries, resulting in premature atherosclerosis.
G5 and G8 are both glycoproteins (11). If recombinant G5 or G8 is expressed alone in cultured cells, the recombinant protein is retained in the ER and remains sensitive to endoglycosidase (Endo) H, an enzyme that removes only immature, high mannose N-linked glycans that have failed to undergo processing to mature oligosaccharides in the Golgi complex. If both halftransporters are expressed in the same cell, the N-linked sugars become insensitive to Endo H and sensitive to neuraminidase, which cleaves sialic acid residues added in the trans-Golgi complex. As G5 and G8 transit through the Golgi complex to the cell surface, the changes in sugar content increase the apparent molecular mass of each protein on SDS-PAGE gels (11). When both proteins are expressed in polarized hepatocytes (WIF-B cells), the proteins are transported to the apical domain (11), but the locations of G5 and G8 within enterocytes and hepatocytes in vivo remain to be determined.
It is generally accepted that ABC half-transporters function as homo-or heterodimers. The observations that co-expression of G5 and G8 is required for either protein to reach the cell surface and that the two proteins co-immunoprecipitate when co-expressed in cultured cells (11), are consistent with these two members of the G subfamily functioning as a heterodimer. It remains possible that homodimers of either G5 or G8 are functional. Differentially epitope-tagged variants of G8 coprecipitate, but the putative dimers fail to exit the ER (11). Additional evidence that G5 and G8 must heterodimerize to transport sterols comes from the molecular analysis of sitosterolemia. Mutations in either ABCG5 or ABCG8 cause a disease of clinically identical phenotype (3,12). Moreover, the two genes are located in a head-to-head orientation on chromosome 2; this orientation is typical for genes that encode subunits of functional complexes that share common regulatory sequences (viz. divergent promoters) (13,14). ABCG5 and ABCG8 have similar tissue-and cell-specific patterns of expression and are coordinately regulated (3,15), which is consistent with these genes sharing common regulatory elements.
Some members of the G subfamily of ABC transporters function as homodimers (ABCG2), whereas others heterodimerize with more than one subfamily member to transport different substrates. In Drosophila, eye color is determined by three members of the G subfamily of ABC half-transporters: white, brown, and scarlet (16). The white gene product heterodimerizes with either brown or scarlet to transport amino acids into pigment granules (17)(18)(19). Given this precedent, it is possible that G5 or G8 form functional heterodimers with other G subfamily members expressed in the liver and intestine.
ABCG2, which was originally identified in cells resistant to chemotherapeutic drugs, is expressed on the apical (lumenal) surface of enterocytes and hepatocytes (20,21). G2 plays a crucial role in the excretion of chlorophyll metabolites into bile (22). Although G2 forms a homodimer and can transport substrates in the absence of any other subfamily member in cultured cells (23,24), it is possible that this protein heterodimerizes with other G subfamily members in vivo.
ABCG1 has been implicated in cholesterol efflux from macrophages (25) and is expressed in the liver, as well as other tissues (26,27). ABCG4 is expressed predominantly in the brain, eye, spleen, and bone marrow and is expressed at low levels in the liver (28,29). The physiological functions and dimerization partners of G1 and G4 are not known.
The present study confirms that cell surface G5 and G8 are localized to the apical membranes of enterocytes and hepatocytes in vivo. Co-expression of G1, G2, and G4 fail to support trafficking of G5 or G8 beyond the ER in cultured cells. In addition, G5 and G8 require co-expression for trafficking and biliary cholesterol excretion in G5G8 null mice.

MATERIALS AND METHODS
Cloning of ABCG Subfamily Members-Epitope-tagged mouse G5 (mG5-Myc) and mouse G8 (mG8-HA) cDNA expression constructs were described previously (11). Human ABCG1-Myc/His and ABCG2 expression plasmids were kindly provided by Peter Edwards (UCLA) and Daniel Garry (University of Texas Southwestern Medical Center), respectively. The human ABCG4 cDNA was amplified from an IMAGE clone (identification number 1537140, accession number AA912226) using two oligonucleotides (5Ј-CTGGCTAGCATGGCGGAGAAGGCG-3Ј and 5Ј-AACTCGAGTCAGCGGCCGCTTCTCTCTGACTT-3Ј) and Pfu polymerase (Stratagene, La Jolla, CA). The ABCG4 cDNA was cloned into the NheI and XhoI sites of pcDNA3.1/Zeo. A 93-bp fragment containing three copies of an immunogenic epitope (YPYDVPDYAG) from hemagluttinin (HA) (30) was inserted into a NotI site within the 3Ј oligonucleotide.
Human ABCG5 and ABCG8 cDNAs were amplified by PCR from a human liver cDNA pool (BD Biosciences Clontech, Palo Alto, CA) using Pfu polymerase and oligonucleotides complementary to the 5Ј and 3Ј ends of the coding region (ABCG5 forward, 5Ј-CTGGCTAGCATGGTG-ACCTCTATCT, and reverse, 5Ј-AACTCGAGTCAGCGGCCGCTCCTG-CTAATGAG; and ABCG8 forward, 5Ј-CTGGCTAGCATGGCCGG-GAAGGCGGCA, and reverse, 5Ј-AACTCGAGTCAGCGGCCGCTCCA-GTCTTGACT). ABCG5 and ABCG8 cDNAs were cloned into the NheI/XhoI sites of pcDNA3.1(ϩ) and pcDNA3.1/Zeo(ϩ), respectively. Attempts to ligate the complete ABCG5 cDNA into pcDNA3.1(ϩ) failed repeatedly. Therefore, the complete cDNA was amplified and digested with NheI and BamHI, BamHI and EcoRI, or EcoRI and XhoI. The resultant fragments were subcloned in series into pcDNA3.1(ϩ) to form the ABCG5 cDNA expression construct. The 3Ј oligonucleotide used for the initial amplification also contained a NotI site upstream of the internal stop codon and XhoI site. A 105-bp fragment encoding three copies of a Myc epitope (EQKLISEEDLN) (11) was inserted into the ABCG5 cDNA clone at the NotI site. In addition, a cDNA encoding three copies of the FLAG epitope (DYKDHD) was generated using overlapping oligonucleotides (5Ј-TAGCGGCCGCATGGATTACAAGGAT-CACGA, 5Ј-TGATCCTTGTAATCTCCGTCGTGATCCTTG, 5Ј-TTAC-AAGGATCACGACGGAGATTACAAGGA, and 5Ј-TAGCGGCCGC-CGTCGTGATCCTTGTAATCT), amplified by PCR, and inserted into the NotI site of the ABCG5 expression construct.
Development of Monoclonal Antibody to ABCG8 -G5G8 Ϫ/Ϫ mice were given a single intrasplenic injection (100 g in phosphate-buffered saline; day 0) of a cDNA encoding full-length mABCG8 in the pcDNA 3.1/Zeo mammalian expression plasmid, followed by two intramuscular injections of the same cDNA (100 g each; days 28 and 42) and four additional injections (300 g each, intraperitoneally; days 56, 70, 84, and 98) of a recombinant fragment (amino acids 1-350) of mouse G8 protein. The fragment of mouse G8 was expressed in Escherichia coli using the pET28a expression system. Serum was collected from mice on day 84 and screened for immunoreactivity by enzyme-linked immunosorbent assay and by immunoblotting using the recombinant G8 protein. Once a positive response was established, the recombinant protein was administered 3 days prior to removal of the spleen. A 5:1 ratio of spleen/myeloma (SP2/0-IL 6) cells were fused with 50% (w/v) polyethylene glycol and plated in 96-well plates (31). The clones were screened by enzyme-linked immunosorbent assay and immunoblot analysis.
Isolation of Plasma Membrane Sheets-Plasma membrane sheets were isolated from mouse liver using an established procedure developed for rats with minor modifications (32). Briefly, five mice (ϳ30 g) were anesthetized with halothane and killed by guillotine. The livers were dissected and perfused via the portal vein with ice cold phosphatebuffered saline. The livers were pooled, weighed and minced on ice. The subsequent steps of the procedure were performed exactly as described (32).
Immunofluorescence Microscopy-Mice were anesthetized with 0.8 mg of pentobarbital and perfused via the left ventricle with Hanks' balanced salt solution (40 ml) and 40 ml of Fix-1 solution (3% paraformaldehyde, 3 mM picric acid, 4 mM KCl, and 2 mM MgCl 2 in phosphatebuffered saline, pH 7.5). The liver and intestines were dissected and placed in Fix-1 for 2 h. Fix-1 solution was aspirated, and the samples were incubated with two changes of Fix-2 solution (60% methanol, 8% acetic acid, 2% trichloroacetic acid, 30% methylchloroform) for 12 h each. Fix-2 solution was aspirated, and the samples were incubated with three changes of 70% ethanol for 30 min each. The samples were transferred to 100% ethanol for 16 h, dehydrated in xylene, and embedded in paraffin. The sections (10 M) were cut, floated onto positively charged slides, and post-fixed in Ϫ20°C methanol for 5 min. Incubation with primary and secondary antibodies as well as confocal imaging were performed as described previously (11).
Transfections and Cell Lysates-The CHO-K1 cells were seeded (1 ϫ 10 5 cells/dish) in 35-mm dishes. The expression plasmids (1 g/dish) were transiently transfected into the CHO-K1 cells using FuGene (6 l of FuGene/g of DNA) according to the manufacturer's protocol. The cell lysates were prepared 48 h after transfection. The cells were washed twice in phosphate-buffered saline (pH 7.4), incubated in 0.3 ml of Triton lysis buffer (50 mM Tris, 80 mM NaCl, 2 mM CaCl 2 , 1% (v/v) Triton X-100, 1 mM phenylmethylsulfonyl fluoride, 10 g/ml aprotinin, 10 g/ml leupeptin, and 5 g/ml pepstatin A, pH 8.0) for 30 min at 4°C and scraped with a rubber policeman. The cells and lysis buffer were transferred to a 1.5-ml tube and centrifuged for 15 min (15,000 ϫ g, 4°C). The supernatants were transferred to a new 1.5-ml tube and subjected to SDS-PAGE and immunoblot analysis.
Immunoprecipitations-Immunoprecipitations were conducted essentially as described (11) with the following modifications. 48 h after transfection, the lysates were prepared in lysis buffer (50 mM HEPES, 100 mM NaCl, 1.5 mM MgCl 2 ) containing protease inhibitors and 0.5% (w/v) digitonin (Calbiochem, La Jolla, CA). The lysates were incubated for 16 h at 4°C on a rotator in the presence of the indicated antibodies and protein A agarose. The pellets were washed in digitonin lysis buffer three times for 10 min at 4°C on a rotator. Analysis of several deter-gents indicated that G5/G8 heterodimers were more stable in digitonin than in Nonidet P-40, Triton or CHAPS. 2 SDS-PAGE and Immunoblot Analysis-The protein concentrations of cell lysates and fractions were determined using the Bio-Rad Dc assay according to the manufacturer's protocol. Protein sample buffer was added to a final concentration of 1ϫ, and the samples were heated to 95°C for 5 min. Proteins were size-fractionated on 10% SDS-polyacrylamide gels at 50 mA and subsequently transferred to nitrocellulose membranes at 50 V for 2 h. The membranes were incubated in buffer A (20 mM Tris, pH 7.6, 137 mM NaCl, 0.5% Tween 20, 5% milk) for 60 min at 22°C prior to the addition of polyclonal antibodies directed against calnexin (Stressgen, Victoria, Canada), the Myc (A14) or HA (Y11) epitopes (Santa Cruz Biotechnology, Inc., Santa Cruz, CA) or monoclonal antibodies directed against the FLAG epitope (M2; Sigma) or ABCG2 (BXP-21; Signet, Dedham, MA). Primary antibodies were diluted in buffer A and incubated with membranes for 60 min at 22°C. The membranes were washed three times for 5 min in buffer B (20 mM Tris, pH 7.6, 137 mM NaCl, 0.5% Tween 20, 0.2% nonfat milk). Horseradish peroxidase-conjugated goat anti-rabbit IgG (Sigma) was diluted (1:10,000) and incubated with membranes for 60 min at 22°C. The membranes were washed three times for 5 min in buffer B and visualized using enhanced chemiluminescence (Amersham Biosciences). Protein loading was assessed by visual inspection of Ponceau S-stained membranes.
Adenoviral Injections of G2, G5, and G8 -Recombinant adenoviral vectors containing cDNAs for ABCG2, ABCG5, and ABCG8 were generated by in vitro cre-lox recombination as described previously (11). The viruses were propagated in HEK-293 cells, purified on cesium chloride gradients, and dialyzed in Tris-buffered saline (10 mM Tris, pH 7.5, 150 mM NaCl). The mice were infected with adenovirus (5 ϫ 10 12 particles/kg) by tail vein injection. After 72 h, the mice were anesthetized with halothane and killed by exsanguination. Serum and bile were collected and analyzed for neutral sterol concentrations by gas liquid chromatography and mass spectroscopy as described (7). Liver tissue was snap frozen in liquid nitrogen and stored at Ϫ80°C until processing for membrane preparation, SDS-PAGE, and immunoblot analysis.

RESULTS
Recombinant, epitope-tagged G5 is localized to the apical surface of polarized hepatocytes when co-expressed with G8 (11). To confirm that the glycoproteins were present on the apical surfaces of cells in vivo, immunofluorescence microscopy was used to localize G5 in the intestine of wild-type, G5G8 null (G5G8 Ϫ/Ϫ ), and G5G8 transgenic mice (Fig. 1). No signal was apparent in the small intestine of G5G8 Ϫ/Ϫ mice, demonstrating the specificity of the antibody. In contrast, the apical surface of the villi was stained in the sections from the wild-type mice and in mice expressing ϳ14 copies of a human fragment containing both ABCG5 and ABCG8 (6) (G5G8 Tg ). Punctate intracellular signal was also seen in the cytoplasm of enterocytes. When the same antibody was used to examine the liver, the signal was too low in wild-type animals to convincingly localize the protein (data not shown). Therefore, we examined the localization of ABCG8 in liver membranes from wild-type and G5G8 Tg mice using immunofluorescence microscopy (Fig.  2). Antibodies directed against apical (APN), basolateral (HA321), and tight junction markers were used to define domains within the plasma membrane sheets (Fig. 2A). The localization of ABCG8 was compared with the tight junction protein ZO-1, which defines the border between apical and basolateral domains (Fig. 2B) (33). The signal for human ABCG8 was essentially circumscribed by the ZO-1 signal, indicating that cell surface ABCG8 is highly enriched, if not present exclusively on the apical membranes of hepatocytes in vivo. An identical pattern was observed for mouse ABCG5 (data not shown).
To address the possibility that G5 or G8 form functional complexes with other ABCG subfamily members, we first compared the sequences of the human ABCG transporters by multiple sequence alignment using Clustal W within the Biology Workbench Suite (San Diego Supercomputer Center; Fig. 3). Little sequence conservation was apparent between the proteins except in the nucleotide binding fold, including the Walker A motif (Fig. 3A), the Walker B motif (Fig. 3B), and the signature sequence (Fig. 3C). In addition, three other amino acid residues that are highly conserved within the ABC transporter subfamily were conserved in these five subfamily members: the glutamine in the Q loop, the aspartate in the D loop,  2. Immunofluorescence microscopy of human G8 in plasma membranes isolated from the livers of wild-type and G5G8 Tg mice. Hepatic plasma membranes from wild-type (WT) and G5G8 Tg mice were isolated, spotted on 24-mm 2 coverslips, fixed in Ϫ20°C methanol, and processed for immunofluorescence microscopy. A, domains from the isolated plasma membranes were determined using antibodies directed against APN (apical), HA321(basolateral), and ZO-1 (tight junctions). B, to determine the localization of human G8 in plasma membranes of G5G8 Tg mice, fixed plasma membrane sheets from wild-type and G5G8 Tg mice were incubated with a rabbit antihuman G8 polyclonal antibody and a monoclonal antibody directed against the tight junctional marker ZO-1.
FIG. 1. Immunofluorescence microscopy of mouse G5 in the intestine of G5G8 ؊/؊ , wild-type, and G5G8 Tg mice. The intestines were fixed and processed for immunofluorescence microscopy as described under "Materials and Methods." Tissue sections were incubated with a polyclonal antibody directed against the N terminus of mouse G5 (4591) and an Alexa 488-conjugated secondary antibody. and the histidine in the so-called switch region (2). The relative sequence identity and evolutionary distance between subfamily members was determined by ProtDist (Table I) and illustrated in a rooted phylogenetic tree (Fig. 3B). ABCG1 and ABCG4 are more closely related to each other, but the other members of the subfamily are equally divergent. Surprisingly, ABCG5 and ABCG8 are only slightly more closely related to one another than to any of the other three members within the subfamily.
To determine whether other G subfamily members interact with G5 or G8 in cells, we performed experiments with the expression constructs shown schematically in Fig. 3C. Because antibodies for G1 and G4 were not available, epitope tags were added to the C termini of these proteins. Placement of epitope tags of similar length at the C termini of G5 and G8 does not interfere with formation of the G5⅐G8 complex or trafficking of the heterodimer to the apical membrane (11).
G5 and G8 are glycoproteins; if either protein is expressed in cells, it remains sensitive to Endo H, indicating the protein is retained in the ER (11). When co-expressed, the carbohydrate chains of G5 and G8 become resistant to Endo H but sensitive to neuraminidase, indicating that the proteins have moved to the trans-Golgi complex (11). We used the characteristic in-creases in apparent molecular mass for G5 and G8 that accompany these post-translational modifications to determine whether other members of the G subfamily could support trafficking of either G5 or G8 in transiently transfected CHO-K1 cells. G5 and G8 were co-expressed with each of the other three G subfamily members: G1, G2, and G4 (Fig. 4A). The mature glycosylated forms of G5 were present only when G5 was expressed in the presence of human or mouse G8, but not when expressed alone or co-expressed with G1, G2, or G4. Similarly, G8 was retained in the ER unless it was co-expressed with human or mouse G5. Nitrocellulose filters from these experiments were stripped, and immunoblotting using antibodies to protein (G2) or to epitope tags (G1 and G4) confirmed the expression of the other G subfamily members. These data indicate that none of the other G subfamily members supported movement of G5 or G8 out of the ER in these cells. Thus, for G5 and G8 to escape the ER, the two proteins must be co-expressed, and human and mouse G5 and G8 are sufficiently similar in sequence that the interspecies half-transporters can heterodimerize with each other.
To determine whether any of the other ABCG subfamily members dimerize with G5 or G8, cell lysates from the experiment described above (Fig. 4B) were prepared in buffer con- FIG. 3. Alignment of the human G subfamily members, relative evolutionary distance, and expression constructs. A, the inferred amino acid sequence for each G subfamily member was aligned using Clustal W within the Biology Workbench Suite (San Diego Supercomputer Center. Walker A, B, and C motifs are underlined, and identical residues are boxed. The other residues conserved in other ABC family members are in bold type (see text). B, the relative evolutionary distance between subfamily members was determined by ProtDist and is illustrated in a rooted phylogenetic tree. C, schematic diagrams of expression constructs containing epitope tags. H, histidine. taining 0.5% digitonin. Monoclonal antibodies against an irrelevant epitope (2001), G1 (9E10, Myc), G2 (BXP21), G4 (12CA5, HA), G5 (M2, FLAG), or G8 (1B10A5) were added to lysates, which were then incubated overnight at 4°C. The antibodybound proteins were captured with protein A-agarose, and the immunoprecipitates (5%) and supernatants (1%) were subjected to SDS-PAGE and immunoblot analysis to detect G5 or G8 (Fig. 4B). Immunoblot analysis indicated that all of the G subfamily members were quantitatively precipitated by their respective antibodies (data not shown). A small amount of the immature form of G5 co-precipitated with G1, G2, and G4. Thus, any complex formed between G5 and G1, G2, or G4 in these cells was retained in the ER. Only when G8 was co-expressed with G5 did any of the mature forms of G5 coimmunoprecipitate. A small proportion of the immature form of G8 co-immunoprecipitated with G4 but not with G1 or G2. Only when G5 was expressed in these cells were the mature forms of G8 present in the pellet, indicating that putative G8⅐G4 complexes were confined to the ER.
Although G1, G2, and G4 were unable to support the trafficking of G5 and G8 out of the ER in vitro, such complexes might form in vivo. To determine whether any of the ABCG subfamily members or any other endogenous proteins could support the trafficking of G5 or G8, adenoviruses expressing recombinant G5, G8, or both proteins were used to infect G5G8 Ϫ/Ϫ mice (7) (Fig. 5). The functional activity of the com-  plexes was assessed by measuring the levels of biliary sterols.
Expression of G5 or G8 alone did not increase the concentration of biliary cholesterol or any other neutral sterol detectable by gas chromatography, compared with G5G8 Ϫ/Ϫ mice (Fig.  5A). In contrast, co-expression of G5 and G8 increased biliary concentrations of cholesterol greater than 10-fold relative to G5G8 Ϫ/Ϫ mice and 2.5-fold relative to wild-type animals. Immunoblot analysis of liver membranes revealed that fully processed forms of G5 and G8 were observed only when the two half-transporters were co-expressed (Fig. 5B). These data showed that endogenous proteins, such as other ABCG subfamily members, could not support measurable trafficking of G5 or G8 out of the ER in vivo and indicated that co-expression of ABCG5 and ABCG8 are required to promote excretion of cholesterol into bile. It remains possible that G5 or G8 homodimerizes or heterodimerizes with another subfamily member and functions in a manner not assayed by these studies. However, these potential dimers would be confined to the ER because expression of G5 or G8 individually failed to allow either protein to reach the trans-Golgi complex.
Although expression of G5 or G8 alone failed to increase biliary cholesterol concentrations in knockout mice, it is possible that the high levels of expression of G5 or G8 may overwhelm the ability of the cell to process the heterodimers. We used reverse transcriptase-PCR to examine the expression levels of G1, G2, and G4 relative to G5 and G8 in the livers of these animals and found the levels of G1 and G4 mRNAs to be very low (data not shown). Moreover, Hoekstra et al. (27) recently reported that G1 is expressed predominantly in the Kupffer cells and endothelial cells but not in hepatocytes in mouse liver. In contrast to G1 and G4, G2 mRNA is expressed abundantly in the liver and has been localized to the bile canalicular membrane using immunocytochemistry (20). To test the possibility that G2 may heterodimerize with either G5 or G8 or may transport sterols into bile as a homodimer, mice deficient in G5 and G8 were infected with adenoviruses expressing G2 alone or in the presence of G5 and G8. Co-expression of G5 and G8 was used as a positive control in the experiment (Fig. 6). Expression of G2 alone or in the presence of G5 or G8 failed to increase biliary cholesterol concentrations (Fig. 6A). No increase was observed for sitosterol, campesterol, or stigmasterol. To ensure that the G2 used in this experiment had been processed when expressed in CHO-K1 cells and mouse liver, cell lysates were subjected to glycosidase treatment (Fig. 6C). ABCG2 was sensitive to N-glycanase and resistant to Endo H treatment in both cultured cells and in the liver, indicating that the protein had folded properly and had trafficked beyond the ER. Moreover, the recombinant G2 protein was sensitive to neuraminidase and thus has acquired sialic acid residues in the Golgi. Thus, G2 did not require expression of other ABCG proteins to exit the ER but failed to support trafficking of G5 or G8 (Fig. 6B) out of the ER. DISCUSSION In the present study, immunofluorescence microscopy provided the first direct evidence that G5 and G8 are located predominantly on the apical membranes of enterocytes and hepatocytes in vivo. The other major findings of this study are that co-expression of G5 and G8 is required for either protein to be transported beyond the ER in vivo and to promote transport of neutral sterols from hepatocytes into bile. Expression of other ABCG subfamily members (G1, G2, or G4) in the same cells as G5 and G8 failed to support movement of either protein out of the ER. All of the complexes that formed between G5 and G8 and any other ABCG subfamily members were retained in the ER. These data indicate that G5 and G8 are obligate heterodimers and are dependent on one another for transport out of the ER and for the excretion of hepatic sterols into bile. Mice lacking G5 and G8 were infected with adenoviruses encoding ␤-galactosidase (␤-gal), mouse G5 (mG5), mouse G8 (mG8), or G5 and G8 together (G5ϩG8). Mice infected with ␤-galactosidase, G5, or G8 also received an equivalent amount of empty virus. Bile was collected following a 4-h fast. A, biliary cholesterol concentrations were determined by gas chromatography-mass spectrometry. B, expression of G5 and G8 was confirmed by SDS-PAGE and immunoblot analysis of liver membrane fractions using antibodies to mouse G5 and G8. Equal sample loading was confirmed by the presence of the ER resident protein calnexin. wt, wild type.
The G subfamily is one of two ABC transporter subfamilies in which members contain only one nucleotide binding fold and one transmembrane domain. As a consequence, these ABC half-transporters must either homodimerize or heterodimerize to form a functional complex. In humans, the number of G subfamily members is relatively small compared with some plants that have 29 transporters of this class (34). A potential way to augment the repertoire of substrates that each ABCG subfamily member can transport is to form heterodimers with more than one partner. Precedent for use of this strategy is found in Drosophila, where White heterodimerizes with Brown to transport precursors of a red pigment and with Scarlet to transport precursors of brown pigment into the pigmentary granule of the eye (16). Because of the differences in substrate specificity of the White/Brown and White/Scarlet complexes, mutations in brown have a different phenotypic effect (brown eyes) than do mutations in either scarlet (red eyes) or white (white eyes).
The ability of White to dimerize with multiple partners raises the interesting possibility that more than one functional coupling occurs between the human ABCG subfamily members, enabling these proteins to transport a broader spectrum of substrates. This is a particularly attractive model given that more than 200 different plant-and animal-derived sterols are present in the diets of omnivores (35). No phenotypic differences have been appreciated in patients who have sitosterolemia caused by mutations in ABCG5 or in ABCG8 (3,12), which is consistent with these half-transporters being monogamous heterodimers. The results of the present study indicate that G5 and G8 are both required for the efficient secretion of sterols into bile and that none of the other ABCG subfamily members will support the transport of G5 or G8 to the cell surface.
Formation of a functional complex between half-transporters would require expression of the two proteins in the same cells. G1 and G4 are expressed at only low levels in the liver. Recently it was reported that G1 mRNA is predominantly expressed in Kupffer cells and endothelial cells rather than in hepatocytes (27). Because no mRNA for G5 or G8 has been convincingly demonstrated to be present in Kupffer cells or any other cells of the macrophage lineage (15), it may be that G1 is not expressed in the same cells as G5 or G8. G4 is expressed at highest levels in the human brain and at much lower levels in the liver and small intestine (28,36), but the cell type-specific pattern of expression of G4 in these tissues has not been determined. Our studies demonstrate that if G4/G5 and G4/G8 heterodimers form in hepatocytes or enterocytes, they fail to exit the ER.
G2, in contrast to G1 or G4, is expressed at appreciable levels in hepatocytes and enterocytes, where the protein has been localized to the apical surface using immunocytochemistry (20). Together with the results of the present study, these data indicate that G2 shares a common subcellular location with G5 and G8 (Figs. 1 and 2). To test whether G2 may complex with either G5 or G8 and participate in the transport of sterols into bile, we used adenoviruses to express G2 in the absence or presence of G5 or G8 and measured the sterol content of the bile using gas chromatography-mass spectrometry. No increase in cholesterol or any other neutral sterol was seen in any of the mice. Thus, G2 appears not to participate in the efflux of neutral sterols from the liver.
It is striking how little sequence identity or similarity exists between the various members of the ABCG subfamily (Fig. 3). As expected, the greatest sequence similarity is in the Walker A motif (GXXGXGK(S/T)) and in the signature sequence (LS-GGQ) (37). In addition, three other highly conserved amino acids implicated to play critical roles in the folding and function of ABC transporters are also conserved between the human G subfamily members. The glutamine in the Q loop, which has been implicated in coupling ATP hydrolysis with conformational changes in the membrane-spanning region (38); the aspartate in the D loop; and the histidine in the switch region, which is required for substrate transport in other ATPases (39,40), are completely conserved among the subfamily members. The specificity for substrate transport resides in the transmembrane domains. The poor conservation among the transmem-FIG. 6. The effect of G2 expressed alone or in the presence of G5 or G8 on biliary cholesterol, sitosterol and campesterol concentrations in G5G8 ؊/؊ mice. Mice lacking G5 and G8 were infected with an empty virus, human G2, G2 in the presence of mouse G5 (G2ϩG5), G2 in the presence of mouse G8 (G2ϩG8), or G5 and G8 together (G5ϩG8). Bile was collected following a 4-h fast. A, biliary cholesterol (top panel), sitosterol and campesterol (bottom panel) levels were determined by gas chromatography. B, expression of G2, G5, and G8 was confirmed by SDS-PAGE and immunoblot analysis of liver membrane fractions using antibodies to human G2 (BXP21), mouse G5 (4591), and G8 (IB10). C, glycosylation analysis of recombinant G2 in CHO-K1 cells (in vitro) and in the liver of G5G8 Ϫ/Ϫ mice infected with G2 adenovirus (in vivo). The cell lysates and membranes were incubated (37°C, 2 h) in the presence of peptide N-glycosidase F (PNGase F), Endo H, and sialidase. The samples were subjected to SDS-PAGE and immunoblot analysis to determine the relative differences in electrophoretic mobility following enzymatic digestion. brane domains of human ABCG subfamily members is consistent with these regions of the genes having diverged substantially to assume new functions.
Whereas G5 and G8 are obligate heterodimers that mediate the efficient excretion of neutral sterols, the physiological functions and dimerization status of ABCG1 and ABCG4 remain unclear. G1, like G5 and G8 (15), is regulated by LXR (41), and expression of G1 is increased by cholesterol loading of macrophages (25). Suppression of ABCG1 mRNA expression decreases cholesterol efflux from macrophages, implicating ABCG1 in the transport of sterols out of cells (25). Although it has been reported that G4 is regulated by LXR and RXR agonists in cultured monocytes (42), no increase in hepatic G4 mRNA was seen in mice fed a high cholesterol diet or treated with a potent LXR agonist. 2 We found no evidence that G5 or G8 form functional complexes with either G1 or G4 in cultured cells or in the liver.
Are G1 and G4 capable of heterodimerizing with each other? These two members of the ABCG subfamily are the most similar in sequence (Table I), especially within transmembrane domains 2 and 5 (42). When the two proteins were expressed together in CHOK1 cells, they co-immunoprecipitated, suggesting that they might heterodimerize in vivo. 2 However, co-immunoprecipitation does not constitute proof of functional dimerization because both G5 and G8 co-precipitated other G subfamily members, but these complexes were retained in the ER. Further studies will be required to determine whether G1 and G4 form functional heterodimers.
G2 is the most extensively studied member of this subfamily. Expression of G2 in cells devoid of other G subfamily members promotes the efflux of chemotherapeutic agents, Hoechst dye, and sulfated steroids (43,44), which is consistent with G2 functioning as a homodimer. As anticipated, G2 was resistant to Endo H but sensitive to neuraminidase, when expressed alone in cultured cells (Fig. 6C). Kage et al. (24) demonstrated that ABCG2 (70 kDa) forms a 140-kDa homodimer linked by disulfide bonds, which we confirmed in our studies (data not shown). In contrast to G2, the relative sizes of G5 and G8 are unaffected by the presence of reducing agents when subjected to SDS-PAGE, so presumably these proteins are not joined by disulfide linkages. 2 The localization and functional studies of G1, G2, and G4 support our conclusion that G5 and G8 are obligate heterodimers. It remains a formal possibility that G5 or G8 homodimerize or form heterodimers with other members of the G subfamily to transport substrates other than cholesterol; however, the function of such a complex must be confined to the ER.